1

Transactions

Presented By

Jahanzeb Faizan

2

Articles Mobile Computing and Databases-A survey Daniel BarbaraI IEEE Transactions on Knowledge and Data Engineering,Vol 11.

Correctness Criteria for Multilevel Secure Transactions Kenneth P.Smith, Barbara T. Blaustein, Sushil Jajodia

IEEE Transactions on Knowledge and Data Engineering,Vol 8.

Agent-based transaction processing Little, H.; Esterline, A. Southeastcon 2000. Proceedings of the IEEE , 2000

3

Outline Transaction management in Mobile Computing

Model for Multilevel Transactions

Transaction Processing using Agents

•General Architecture•Unique features•Transaction Models

•Background•Architecture•Methodology•Protocols & Interactions

•Background•Standard Properties•Advancements

4

Transaction management in Mobile Computing

Why need Mobile Computing:

Appearance of powerful portable computers

Development of fast reliable network

5

Architecture of General Mobile Environment

Two Entities Mobile units Fixed Host

6

Asymmetry in Communication

Unique Features

Data Dissemination

Impact on transaction Management

Roaming of clients through different cells

•Location Dependent Queries Issue

Impact on kind of interfaces

Frequent Disconnecting Power Limitations Screen Size

7

Transaction Models Escrow Method Walborn & Chrysanthis Generilazation Demarcation Protocol Two Tier Replication Algorithm Certification reports Isolation only transactions

8

Escrow Method Divides the total number of instances of

an item among no. of sites in system.

A transaction can only successfully complete at a site if no. of instances it requires does not exceed available instances at that site.

9

Client

Server1 Server2

Trans

1

Trans 1

Move to S2

It will take decision based on available instances

It only need next operation to be performed

Syn. By 2 Phase Commit Algorithm

Demonstration

10

Walborn & Chrysathis Generilazation

Splits large and complex objects into consistent sets of smaller fragments

Cache a set on Mobile client

Advantage:

Concurrent Operations on a set of Mobile Clients

11

Split Operation on object

1.Selects part of object

2.Establishes consistency conditions

Client Server

Sends request to access object with condition

Part of the object is only accessed by the client

Remaining part available for other clients

Demarcation Protocol

12

Two Tier Replication Algorithm

Tentative transactions at mobile clients on replicated data while they are disconnected

Connectivity

No

Perform tentative transactions on replicated data

Yes

Replicated UpdatesFail

updatesInform the client

13

Two Tier Replication Algorithm (cont’d) Advantage:

Standard way of propagating updates to replicas.

Disadvantage:

May result in unacceptable no. of failed transactions

14

Certification Reports Provides a technique with which broadcast

channel can be used to help mobile clients to do some verification for a transaction running by them and need to be aborted.

15

Algorithm

Client A

Active transactionSends request to server

Certification Report

Client B

Active TransactionListen to CR by broadcast

channel

Compare itemsets of current tran with CR

ok

Not

ok

Trans abort

Server

Send request to server

Server checks again with CR

commit

okNot ok

Aborts

16

Certification Reports (cont’d) Advantage

•Most of the work of verification by clients

•Early abortion of false transaction

17

Isolation Only Transactions(IOT) Strong Consistency can be guaranteed

Without guarantying other transaction properties such as atomicity and durability.

18

Execution of transaction on client

Requires partitioned data accessPending state and waits for validation

Successful validation Unsuccessful validationReintegration of results Resolved manually or

automatically

Commit on server

Do not Requires partitioned data

accesscommitted

Results visible on server

Isolation Only Transactions(IOT)-cont’d

19

Model for Multilevel Transactions

What is Multilevel Security Level:

•Users could only access data on which they have security rights

•No “writing down”

In same session,reading from one level and writing to a lower level.

20

Current Multilevel Secure DBMS They implement multi security level. No writing down.

Solution :

MUSET (Multi level Secure Transactions) Project

21

MUSET

Layered Architecture of MUSET:

MLS DBMS

Single Level Applications

Multilevel Applications

Single Level DBMS Processes

Multilevel DBMS Processes

Interconnection

D-DBMS Appl. Layer

MUSET

22

Operation

Site 1X

Stores low data

Site 1Y,Z

Stores high data

Let site 2 issue following ML site transparent transaction

Y:=1

X:=X+2

Z:=X+Y

23

MUSET Analysis

High :W(Y)Low : R(X)Low: W(X)High: R(X)High: R(Y)High: W(Z)

Low Labeled X data item is labeled high

High Process must read X in order to

write high data item Z

Y:=1

X:=X+2

Z:=X+Y

24

Methodology

MUSET divides the operations into single level sets or sections

Low R(X)W(X)

High W(X)R(X)R(Y)W(Z)

Site 1 Site 2

Coordination according to original transaction

25

Correctness Criteria

Goal Execute each section while achieving atomic,consistent,isolated and

secure execution of the original transaction on single site.

Criteria• A-correctness Fully atomic. Each commit or none• C-correctness Conflict equivalent to original transaction.• I-correctness Conflict equivalent to a serial ordering. • S-correctness No interference.

Drawback:

• Except security all are achievable in multilevel transactions.

26

Protocols Multilevel Transaction Schedule(STi) Total order of operations with ordering relation <STi

such that;

•Events should be precommit,abort,commit operations

•At each level there should be abort or commit ,but not both•Commit and abort comes after all other events

•Precommit, if exists should come before commit or abort

•Execution must preserve the order of conflicting operations with in a level.•If read at high level follows a lower level write of same data,read must follow the commit of write.

27

Multilevel Schedule ST over a set of multilevel transactions T1,T2,--,Tn is the

ordering of all operations in ST1, ST2, ST3,... STn, with ordering relation <ST,such that it preserve the ordering of all operations in each STi

28

Multilevel Execution Protocol (P)

Transformation of a set of transactions T into a schedule ST.P(T)=ST

A protocol P is A-correct,I-correct,C-correct and S-correct iff for every set T of multilevel transactions, P(T) is A-correct,I-correct,C-correct and S-correct respectively.

29

Interactions of Correctness Criteria

Atomicity and Security No protocol can be both A- and S-correct.

Proof:

ST1

ST2

ST3

ST4

Execute and Precommit

A-Correct

pre

com

mit

30

Atomicity and Security (cont’d)

Precommitting in ascending order is a timing channel,because commit of lower operations is being delayed due to the execution of higher level operations.

“So it is guaranteeing the A-correctness ,but not s-correctness”

31

Consistency & Security

Scenario: High level read R(X) reads the value x1 for x and then

low level write W(X) writes value x2.

This is violating S-correctness.So reorder.But now wrong version of value would be read by high level.This is conflicting C-correctness.

Server

R(X), X1 W(X), X2

32

Consistency & Security (cont’d)

Solution: USE OF CACHING The value x1 is cached before updating to x2. So instead

of reading x2,high level read would read x1 from cache.

cache

server

W(X) ,x2

Cache x ,x1

R(X), X1

33

Isolation & Security Isolation by 2PL:

2PL Locks:

Intralevel locks: All read and write locks within a level.

Interlevel locks: Read locks covering a high read of a low datum.

Intralevel lockson each security level

Interlevel lockson “read down” operation.

Security hazard!In High reads,

low writes locks are prohibited

34

Isolation & Security (cont’d) Solution: SOFT LOCKS High read lock is broken when low level

transaction requires lock.

•Unlock can occur unpredictably either before or after execution of read ,which is a threat to I-correctness.

•So, no protocol that uses 2PL can be I- and S-Correct

35

Execution Protocols

Low Ready wait 2PL (ACIS` correct) Low First Multiversion Time stamping Ordering(A`CIS correct) Low First Hybrid Multiversing (A`CI`S correct)---used by Oracle

36

Transaction Processing using Agents

What is a Agent: Autonomous entity that can meet required goal without

the assistance of a human.

Focus of Paper: Interaction of agents in a multi agent system by

applying transaction abstraction to ensure integrity

37

Properties ACID properties are not satisfied in dealing

with agents.

Isolation is violated in two agents acting on same resource.

Atomicity is violated when attempting to get a quote for an airline reservation among different companies,because they will not let you run commit protocols.

38

New Properties Spheres of Control Socs attempt to contain the effects of an action as long

as there might be a necessity to undo them

Process atomicity All or none

Process control control given to process so that other process cannot modify it

Process commitmentidentifies a function which determines the modified value of each data item.

39

Spheres of Commitment

Once an agent makes an commitment it is obligated to keep that commitment

Under presence of a witness,governing all parties at events

The committee (bearer) could discharge from it

The committer accepts request of discharge

The committer could discharge the committee

40

Society Witness Top level Committer middle level Committeebottom level

41

Aglets They are used to implement the agents

because they allow Freedom to create mobile and collaborative

agents which has the ability to transport themselves throughout the network to meet specific goal.

42

Future Work Study of real cases. Which properties can be effectively ignored or

relaxed Implementation of protocols Development of protocols capable of executing

transactions over a set of DBMSs using significantly different approaches to features such as concurrency control.

43

Conclusion The inherent limitations of mobile computing

systems present a challenge to the traditional problems of DBMS

Security requirements of multilevel transactions conflict with other properties.

ACID traditional properties can be transformed to SOCS with commitment rules to effectively perform transactions ,by using agents.